Permafrost thaw subsidence monitoring with InSAR time series
Repeat-pass InSAR time series detect millimetre-scale vertical displacement in Arctic permafrost terrain, separating seasonal frost heave from irreversible thaw settlement to warn of infrastructure risk and carbon-release acceleration.
Sensors
- ALOS-2 PALSAR-2 (L-band, JAXA): L-band (1.27 GHz) penetrates vegetation canopy and maintains coherence across freeze-thaw cycles far better than C-band. Single-look complex resolution approximately 3 m (spotlight) to 10 m (stripmap); 14-day repeat cycle. The longer wavelength (24 cm) reduces phase decorrelation over tundra shrub, making it the preferred sensor for vegetated Arctic terrain.
- Sentinel-1 A/B C-band IW (ESA/Copernicus): C-band (5.4 GHz) at 5 x 20 m resolution in Interferometric Wide Swath mode, 250 km swath, 6-day repeat (12-day with one satellite). Free archive from 2014. Coherence degrades rapidly over dense tundra vegetation and during the active thaw season, but performs well on bare ground, gravel pads and infrastructure corridors. The dense time series is the main advantage for SBAS processing.
- NISAR (NASA/ISRO, forthcoming): Dual-band L+S system scheduled for launch in 2024-2025. L-band (24 cm wavelength) and S-band (9 cm) simultaneously, 12-day repeat, global coverage including high-latitude Arctic. Expected to provide the most coherent vegetation-covered permafrost observations yet from orbit, with co-registered bands enabling direct comparison of penetration depth effects.
- TanDEM-X (DLR): Bistatic X-band pair used primarily to generate a reference DEM at 12 m posting and better than 2 m vertical accuracy over flat terrain. Not used for displacement time series directly, but essential as the external DEM to remove topographic phase from Sentinel-1 and ALOS-2 interferograms, and to detect long-baseline elevation change by differencing against earlier SRTM or ICESat-2 data.
What a sinking centimetre actually means
Permafrost occupies roughly a quarter of the Northern Hemisphere land surface. The top layer, the active layer, freezes each winter and thaws each summer. Below it, ground ice can be thousands of years old. When the active layer deepens year on year, or when ice-rich permafrost degrades, the ground surface subsides. A few centimetres per year sounds trivial. On a pipeline support, a runway, or a building foundation, it is not.
The physics matters for remote sensing. Subsidence from true ice loss is irreversible and cumulative. Seasonal frost heave, where the ground rises in autumn as pore water freezes and falls again in spring, is cyclic and roughly symmetric. A single interferogram cannot tell them apart. A multi-year time series can, because the cumulative trend separates from the seasonal oscillation once you have enough acquisitions to fit both components simultaneously.
How InSAR measures displacement nobody can walk to
Repeat-pass InSAR compares the phase of radar backscatter from two passes over the same scene. Any change in the slant-range distance between satellite and ground shifts the phase by a known fraction of the wavelength. For L-band (24 cm), one full phase cycle corresponds to 12 cm of line-of-sight displacement. For C-band (5.6 cm), the same cycle represents about 2.8 cm. C-band is more sensitive to small displacements in principle, but it loses coherence faster over vegetated or seasonally disturbed surfaces.
The standard processing chain for permafrost work uses either SBAS (Small Baseline Subset) or PS-InSAR (Persistent Scatterer). SBAS stacks many short-baseline interferogram pairs and inverts the network to produce a displacement time series at each coherent pixel. PS-InSAR identifies point-like stable scatterers, typically infrastructure, rocks or gravel, and tracks their phase history with millimetre precision. In practice, permafrost terrain outside infrastructure corridors rarely has enough persistent scatterers for PS-InSAR to dominate; SBAS on L-band is the workhorse.
The coherence problem, stated plainly
Coherence is the correlation between two SAR acquisitions. It falls toward zero when the surface changes between passes, whether from vegetation growth, snow accumulation, soil moisture shifts, or freeze-thaw state change. Over Arctic tundra in summer, C-band coherence over vegetated ground can drop below 0.3 within a single 6-day Sentinel-1 interval. Below roughly 0.3 to 0.4, phase measurements become dominated by noise and cannot be reliably unwrapped.
Phase unwrapping is the step that converts the ambiguous wrapped phase (which cycles between negative pi and positive pi) into a continuous displacement measurement. Where coherence is low, unwrapping algorithms introduce errors that propagate spatially and can produce artefacts that look like real deformation signals. This is not a solvable problem with better software alone; it reflects a physical limit of the wavelength relative to the surface change rate. L-band mitigates it substantially. Some published studies over Siberian and Alaskan permafrost report coherent ALOS PALSAR L-band interferograms over intervals of 46 to 92 days across vegetated tundra where C-band fails entirely.
Atmospheric phase delay is the other major error source. Water vapour in the troposphere adds a spatially variable phase signal that can mimic centimetres of deformation. Over the Arctic, the tropospheric delay is generally lower than in humid mid-latitudes, but it is not negligible. Correction uses ERA5 reanalysis fields or GACOS (Generic Atmospheric Correction Online Service) tropospheric delay maps, both of which reduce residual atmospheric noise to roughly 1 to 2 cm RMS in favourable conditions.
Separating heave from settlement: the seasonal decomposition
A displacement time series over permafrost terrain typically shows a sawtooth pattern: rapid subsidence through the summer thaw season, partial recovery as the ground refreezes in autumn, then a small net loss that accumulates year on year. The amplitude of the seasonal cycle is proportional to active-layer thickness and soil ice content. The long-term trend is the quantity that matters for infrastructure risk and carbon accounting.
The standard decomposition fits a model with a linear trend plus a sinusoidal seasonal term to each pixel's time series. The residuals reveal episodic events: a thermokarst collapse, a drainage change, or a construction disturbance. Reported detection sensitivities for the linear trend component, given multi-year ALOS PALSAR stacks, are in the range of 5 to 10 mm per year in line-of-sight, which converts to roughly 6 to 12 mm per year in the vertical assuming a typical incidence angle near 35 degrees. That is sufficient to detect the early stages of degradation in ice-rich terrain before visible surface failure occurs.
One honest caveat: InSAR measures line-of-sight displacement, not purely vertical. Horizontal motion from slope creep or solifluction adds a component that is indistinguishable from vertical subsidence in a single-geometry acquisition. Ascending and descending track combinations allow decomposition into vertical and east-west horizontal components, but north-south sensitivity remains poor for near-polar orbits.
Infrastructure corridors versus landscape-scale monitoring
The use case splits into two quite different operational problems. For infrastructure, the interest is in detecting differential settlement across a specific asset, a pipeline support, a runway threshold, a building pad, at centimetre or sub-centimetre precision. Here PS-InSAR on Sentinel-1 or ALOS-2 can work well because the infrastructure itself provides stable scatterers. The Yamal LNG facilities, the Trans-Alaska Pipeline corridor, and several Siberian airports have been studied in the published literature using exactly this approach.
For landscape-scale carbon accounting, the interest shifts to area-averaged subsidence rates across tens of thousands of square kilometres, used as a proxy for ice volume loss and therefore for CO2 and methane release from thawing organic carbon. Here SBAS on L-band is the practical method, with spatial averaging over coherent patches to reduce noise. The resulting maps are not precise enough to attribute carbon flux at the pixel level, but they provide a spatially continuous constraint that point-based field measurements cannot.
Satellize processes both problem types on open-constellation data, with commercial ALOS-2 tasking available under client licence for sites where Sentinel-1 coherence is insufficient. Our Tonga crop-estimation programme is a different domain entirely, but the time-series decomposition methods share common lineage with the seasonal signal separation used in permafrost work.
What the archive cannot tell you
Sentinel-1 coverage of high Arctic latitudes is good, with some areas receiving 6-day revisit from overlapping tracks. But the useful InSAR archive only begins in 2014, and the summer acquisition window in which coherence is adequate for thaw-season monitoring is short, perhaps 10 to 14 weeks per year. That gives roughly 100 to 140 summer acquisitions over a decade, which is enough for trend detection but not for resolving inter-annual variability in active-layer depth with high confidence.
ALOS PALSAR-1 data from 2006 to 2011 extends the L-band archive, but the 46-day repeat cycle of the original PALSAR limits the temporal resolution of seasonal decomposition. NISAR, when operational, will change this substantially by providing 12-day L-band globally, including the high Arctic above 80 degrees north where current coverage is sparse.
Finally, InSAR cannot directly measure active-layer thickness or ground ice content. It measures surface displacement. The physical interpretation requires ancillary data: soil type maps, borehole temperature records, or modelled freeze-thaw depth from products such as the ERA5-Land soil temperature layers. The displacement signal is real and precise; the attribution to specific permafrost processes requires that additional context.
Typical figures
| Spatial resolution (ALOS-2 PALSAR-2 stripmap) | 10 m single-look; typically multi-looked to 20-50 m for InSAR coherence |
| Spatial resolution (Sentinel-1 IW) | 5 x 20 m single-look; 20-40 m after multi-looking |
| Revisit interval | Sentinel-1: 6-12 days; ALOS-2: 14 days; NISAR (forthcoming): 12 days |
| Radar frequency / wavelength | L-band: 1.27 GHz / 24 cm; C-band: 5.4 GHz / 5.6 cm |
| Line-of-sight displacement sensitivity (SBAS stack) | 5-10 mm per year for multi-year L-band stacks; ~3-5 mm per year for dense C-band stacks on coherent surfaces |
| Minimum detectable seasonal amplitude | Approximately 1-2 cm peak-to-trough on coherent bare ground; higher threshold over vegetated tundra |
| Swath width | Sentinel-1 IW: 250 km; ALOS-2 stripmap: 70 km |
| Archive depth | Sentinel-1: from 2014; ALOS PALSAR-1 (L-band precursor): 2006-2011; TanDEM-X DEM: 2010-2015 acquisition period |
| Reference DEM vertical accuracy | TanDEM-X global DEM: better than 2 m absolute vertical error over flat terrain |
| Deliverable formats | GeoTIFF displacement rate maps, NetCDF time-series stacks, vector shapefiles of anomaly zones, PDF engineering reports |
Analytics Satellize can run
| Annual subsidence rate map | SBAS InSAR time-series inversion with linear trend fitting per pixel; atmospheric correction via ERA5 or GACOS | GeoTIFF raster of mm/year line-of-sight velocity with uncertainty layer; updated annually or on request |
| Seasonal heave/settlement decomposition | Sinusoidal plus linear model fit to displacement time series; residuals flagged as episodic events | NetCDF stack with trend, amplitude and phase components; anomaly polygons as GIS layer |
| Infrastructure differential settlement report | PS-InSAR on stable scatterers along asset corridor; ascending/descending combination for vertical/horizontal decomposition | Per-structure displacement table and time-series plots; PDF report with risk-ranked asset list |
| Coherence change detection | Mean coherence mapping across summer acquisition windows; year-on-year coherence loss as proxy for surface disturbance | Annual coherence maps; alert shapefile for areas showing coherence decline exceeding threshold |
| Thermokarst lake expansion monitoring | SAR backscatter thresholding combined with subsidence boundary mapping to delineate active thermokarst margins | Annual polygon update of lake boundaries with associated subsidence rate at margins; GIS layer |
| Multi-sensor vertical displacement fusion | Co-registration of InSAR displacement with ICESat-2 ATL06 elevation change and available GNSS benchmarks for absolute calibration | Calibrated vertical displacement raster; validation statistics against independent benchmarks |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.